to rely on the parsed stanza attributes

This was to fix a bug whereby a full stanza was used to query for 1:1
messages with a full JID when the bare JID is stored.

We now are sure that the attributes we're using to query for duplicates
are the same attributes that get saved for a messages.